Neil Conway created MESOS-5950:
----------------------------------
Summary: Consider request/response for reconciliation, bulk
reconcile
Key: MESOS-5950
URL: https://issues.apache.org/jira/browse/MESOS-5950
Project: Mesos
Issue Type: Improvement
Components: framework api, master
Reporter: Neil Conway
The current task reconciliation API has a few quirks:
1. The master will sometimes use "send nothing" as a way to communicate
information (MESOS-4050), which is very confusing in a distributed system that
might drop messages for other reasons.
2. A framework has no way to determine when the reconciliation results for a
given reconciliation request are "complete". That is, when a framework sends a
reconciliation request, it starts to receive zero or more task status updates
(with {{reason}} set to {{REASON_RECONCILIATION}}). The framework can't easily
determine how many results it should expect to receive.
3. For efficiency (and perhaps to simplify framework logic), it might be easier
to send a batch of task status updates together in a single message, rather
than sending potentially tens of thousands of individual messages.
For #2, arguably a framework shouldn't _need_ to know when it has seen the
"complete" set of results for a reconciliation request. However, supporting a
"request/reply" structure for reconciliation can simplify framework logic,
especially if a framework might have multiple timers/reasons to be doing
reconciliation at the same time.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
